organize chapter 9

keon · Sep 3, 2019 · baec02a · baec02a
1 parent db245db
commit baec02a
Show file tree

Hide file tree

Showing 8 changed files with 805 additions and 1,569 deletions.
diff --git a/09-경쟁하며_학습하는_GAN/README.md b/09-경쟁하며_학습하는_GAN/README.md
@@ -1 +1 @@
-# 경쟁을 통해 성장하는 GAN
+# 경쟁하며 학습하는 GAN
diff --git a/09-경쟁하며_학습하는_GAN/conditional_gan.ipynb b/09-경쟁하며_학습하는_GAN/conditional_gan.ipynb
@@ -24,26 +24,6 @@
     "import numpy as np"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 2,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "<torch._C.Generator at 0x7f600d469bf0>"
-      ]
-     },
-     "execution_count": 2,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "torch.manual_seed(1)    # reproducible"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 3,
@@ -58,7 +38,7 @@
     }
    ],
    "source": [
-    "# Hyper Parameters\n",
+    "# 하이퍼파라미터\n",
     "EPOCHS = 300\n",
     "BATCH_SIZE = 100\n",
     "USE_CUDA = torch.cuda.is_available()\n",
@@ -72,7 +52,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Fashion MNIST digits dataset\n",
+    "# Fashion MNIST 데이터셋\n",
     "trainset = datasets.FashionMNIST('./.data',\n",
     "    train=True,\n",
     "    download=True,\n",
@@ -103,7 +83,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Discriminator\n",
+    "# 생성자 (Generator)\n",
     "D = nn.Sequential(\n",
     "        nn.Linear(784, 256),\n",
     "        nn.LeakyReLU(0.2),\n",
@@ -119,7 +99,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Generator \n",
+    "# 판별자 (Discriminator)\n",
     "G = nn.Sequential(\n",
     "        nn.Linear(64 + 10, 256),\n",
     "        nn.ReLU(),\n",
@@ -135,12 +115,12 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "\n",
-    "# Device setting\n",
+    "# 모델의 가중치를 지정한 장치로 보내기\n",
     "D = D.to(DEVICE)\n",
     "G = G.to(DEVICE)\n",
     "\n",
-    "# Binary cross entropy loss and optimizer\n",
+    "# 이진 크로스 엔트로피 (Binary cross entropy) 오차 함수와\n",
+    "# 생성자와 판별자를 최적화할 Adam 모듈\n",
     "criterion = nn.BCELoss()\n",
     "d_optimizer = optim.Adam(D.parameters(), lr=0.0002)\n",
     "g_optimizer = optim.Adam(G.parameters(), lr=0.0002)"
@@ -1156,16 +1136,13 @@
     "        g_loss.backward()\n",
     "        g_optimizer.step()\n",
     "        \n",
-    "        if (i+1) % 200 == 0:\n",
-    "            print('Epoch [{}/{}], Step [{}/{}], d_loss: {:.4f}, g_loss: {:.4f}, D(x): {:.2f}, D(G(z)): {:.2f}' \n",
-    "                  .format(epoch,\n",
-    "                          EPOCHS,\n",
-    "                          i+1,\n",
-    "                          total_step,\n",
-    "                          d_loss.item(),\n",
-    "                          g_loss.item(), \n",
-    "                          real_score.mean().item(),\n",
-    "                          fake_score.mean().item()))"
+    "    print('Epoch [{}/{}], d_loss: {:.4f}, g_loss: {:.4f}, D(x): {:.2f}, D(G(z)): {:.2f}' \n",
+    "          .format(epoch,\n",
+    "                  EPOCHS,\n",
+    "                  d_loss.item(),\n",
+    "                  g_loss.item(), \n",
+    "                  real_score.mean().item(),\n",
+    "                  fake_score.mean().item()))"
    ]
   },
   {

diff --git a/09-경쟁하며_학습하는_GAN/conditional_gan.py b/09-경쟁하며_학습하는_GAN/conditional_gan.py
@@ -14,18 +14,15 @@
 import numpy as np
 
 
-torch.manual_seed(1)    # reproducible
-
-
-# Hyper Parameters
+# 하이퍼파라미터
 EPOCHS = 300
 BATCH_SIZE = 100
 USE_CUDA = torch.cuda.is_available()
 DEVICE = torch.device("cuda" if USE_CUDA else "cpu")
 print("Using Device:", DEVICE)
 
 
-# Fashion MNIST digits dataset
+# Fashion MNIST 데이터셋
 trainset = datasets.FashionMNIST('./.data',
     train=True,
     download=True,
@@ -44,7 +41,7 @@ def one_hot_embedding(labels, num_classes):
     return y[labels]
 
 
-# Discriminator
+# 생성자 (Generator)
 D = nn.Sequential(
         nn.Linear(784, 256),
         nn.LeakyReLU(0.2),
@@ -54,7 +51,7 @@ def one_hot_embedding(labels, num_classes):
         nn.Sigmoid())
 
 
-# Generator 
+# 판별자 (Discriminator)
 G = nn.Sequential(
         nn.Linear(64 + 10, 256),
         nn.ReLU(),
@@ -64,12 +61,12 @@ def one_hot_embedding(labels, num_classes):
         nn.Tanh())
 
 
-
-# Device setting
+# 모델의 가중치를 지정한 장치로 보내기
 D = D.to(DEVICE)
 G = G.to(DEVICE)
 
-# Binary cross entropy loss and optimizer
+# 이진 크로스 엔트로피 (Binary cross entropy) 오차 함수와
+# 생성자와 판별자를 최적화할 Adam 모듈
 criterion = nn.BCELoss()
 d_optimizer = optim.Adam(D.parameters(), lr=0.0002)
 g_optimizer = optim.Adam(G.parameters(), lr=0.0002)
@@ -118,16 +115,13 @@ def one_hot_embedding(labels, num_classes):
         g_loss.backward()
         g_optimizer.step()
 
-        if (i+1) % 200 == 0:
-            print('Epoch [{}/{}], Step [{}/{}], d_loss: {:.4f}, g_loss: {:.4f}, D(x): {:.2f}, D(G(z)): {:.2f}' 
-                  .format(epoch,
-                          EPOCHS,
-                          i+1,
-                          total_step,
-                          d_loss.item(),
-                          g_loss.item(), 
-                          real_score.mean().item(),
-                          fake_score.mean().item()))
+    print('Epoch [{}/{}], d_loss: {:.4f}, g_loss: {:.4f}, D(x): {:.2f}, D(G(z)): {:.2f}' 
+          .format(epoch,
+                  EPOCHS,
+                  d_loss.item(),
+                  g_loss.item(), 
+                  real_score.mean().item(),
+                  fake_score.mean().item()))
 
 
 for i in range(100):
Original file line number	Diff line number	Diff line change
		@@ -1 +1 @@
		# 경쟁을 통해 성장하는 GAN
		# 경쟁하며 학습하는 GAN